JP7477791B2 - Processing device, processing method, and processing program - Google Patents

Description

The present invention relates to a processing device, a processing method and a processing program.

Operations are carried out to improve services based on evaluations of companies and other organizations, with the evaluation targets being people, organizations, existing systems, etc. that are related to the evaluations. The task of analyzing the evaluations and matching the evaluations with the evaluation targets, as well as the analysis, is generally done manually by humans.

In general, there is a method for expressing the development model of an AI (Artificial Intelligence) service system using ArchiMate (see Non-Patent Document 1). There is also a proposal to visualize the current situation and use it to support the selection of a system reconstruction method or improvement points (see Non-Patent Document 2).

Hiroyoshi Takeuchi and 1 other person, "Development model of AI service system for office work", [online], 2019, 24th Knowledge Distribution Network Study Group of the Japanese Society for Artificial Intelligence, [Retrieved May 20, 2020], Internet <URL: https://jsai.ixsq.nii.ac.jp/ej/index.php?active_action=repository_view_main_item_detail&page_id=13&block_id=23&item_id=9778&item_no=1> Information-technology Promotion Agency, Japan, Technology Headquarters, Software Reliability Center, "User's Guide to Successful System Reconstruction - Reconstruction Risks and Countermeasures Shared by Users and Vendors", [online], 2018, [Retrieved May 20, 2020], Internet <URL: https://www.ipa.go.jp/files/000057294.pdf>

However, none of the documents mention matching evaluations with the objects of evaluation through computer processing.

The present invention has been made in consideration of the above circumstances, and the object of the present invention is to provide technology that can easily match evaluations with the objects to be evaluated.

A processing device according to one embodiment of the present invention includes an acquisition unit that acquires an evaluation of an evaluation target, a similarity calculation unit that calculates the similarity between each relation of an entity in the evaluation target and the evaluation, and an output unit that outputs, of the relations, those relations that have a high similarity to the evaluation.

A processing method according to one aspect of the present invention includes a step in which a computer obtains an evaluation of an evaluation target, a step in which the computer calculates a similarity between each relation of an entity in the evaluation target and the evaluation, and a step in which the computer outputs, from among the relations, a relation that has a high similarity to the evaluation.

One aspect of the present invention is a processing program that causes a computer to function as the above-mentioned processing device.

The present invention provides technology that can easily match evaluations with the objects being evaluated.

FIG. 1 is a diagram illustrating functional blocks of a processing device according to an embodiment of the present invention. FIG. 2 is a diagram illustrating an example of the data structure of evaluation data and the data. FIG. 3 is a diagram illustrating an example of the data structure and data of the model data. FIG. 4 is a diagram for explaining an example of the names of entities used in the model data. FIG. 5 is a diagram for explaining an example of a model configuration of an organization to be evaluated. FIG. 6 is a diagram illustrating an example of the data structure of similarity data and the data. FIG. 7 is a flowchart illustrating the acquisition process performed by the acquisition unit. FIG. 8 is a flowchart illustrating the similarity calculation process performed by the similarity calculation unit. FIG. 9 is a diagram illustrating an example of an output result by the output unit. FIG. 10 is a diagram illustrating functional blocks of an output unit according to the first modified example. FIG. 11 is a diagram illustrating an example of a data structure of EA data. FIG. 12 is a diagram illustrating functional blocks of an output unit according to the second modified example. FIG. 13 is a diagram illustrating a learning unit according to the second modified example. FIG. 14 is a diagram illustrating the hardware configuration of a computer used in the processing device.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings. In the description of the drawings, the same parts are given the same reference numerals and the description will be omitted.

(Processing Equipment)
A processing device 1 according to an embodiment of the present invention shown in Fig. 1 uses computer processing to identify, for an evaluation of the evaluation target organization 3, an entity that corresponds to the evaluation among the entities that constitute the evaluation target organization 3. The evaluation is a meaningful block of data of an evaluation by a user of the evaluation target organization, a product or service provided by the organization, etc., and is expressed in language such as a sentence or term. The evaluation is one comment posted by a user, one word-of-mouth posting, etc. The evaluation data set may also be customer complaint information, needs information, etc. managed by CRM (Customer Relationship Management) held by the evaluation target organization 3.

The processing device 1 includes evaluation data 11, model data 12, similarity data 13, acquisition unit 21, definition unit 22, similarity calculation unit 23, and output unit 24. The evaluation data 11, model data 12, and similarity data 13 are data stored in memory 902 or storage 903. The acquisition unit 21, definition unit 22, similarity calculation unit 23, and output unit 24 are functional units implemented in the processing device 1 by execution of a CPU 901 or a GPU (Graphics Processing Unit) (not shown). Here, the similarity calculation unit 23 may be executed by the GPU when processing neural networks such as Word2vec and Doc2vec.

The evaluation data 11 is a collection of evaluations. When an emotional score is calculated for each evaluation, the evaluation data 11 may correspond to each evaluation with its emotional score, for example as shown in FIG. 2.

The model data 12 is data that defines entity relationships related to the evaluation target organization 3. As shown in FIG. 3, the model data 12 links identifiers of the entities that define each relationship. The identifiers of the entities are associated with the names of the entities, as shown in FIG. 4.

In an embodiment of the present invention, the operation model of the organization 3 to be evaluated is defined in a modeling language called ArchiMate. The organization 3 to be evaluated is divided into multiple entities. The organization to be evaluated is also divided into multiple hierarchies, with one or more entities associated with each hierarchy. The operation model of the organization 3 to be evaluated is defined by relations that link entities selected from each hierarchy. In an embodiment of the present invention, the organization 3 to be evaluated is divided into four layers, an organizational layer, an active structure layer, a behavior layer, and a passive structure layer, as shown in Figure 5, with one or more entities associated with each layer. The relations associate entities selected from each layer.

The model data 12 shown in FIG. 3 is formed corresponding to the model of the organization to be evaluated 3 shown in FIG. 5. The model data 12 models the organization to be evaluated 3 by the relationships of each entity constituting the organization to be evaluated 3. The relationships defined by the model data 12 are defined by a combination of a relationship ID, an organization ID, an active structure ID, a behavior ID, and a passive structure ID. The relationship ID identifies the relationship. The organization ID, the active structure ID, the behavior ID, and the passive structure ID are the IDs of entities selected from the organization layer, the active structure layer, the behavior layer, and the passive structure layer that correspond to the relationship, respectively. Note that in an embodiment of the present invention, the IDs of the entities are numbered so as to be identifiable in each layer.

The relationship ID "2" shown in Figure 3 is associated with organization ID "100000", active structure ID "200000", behavior ID "200000", and passive structure ID "200000". The organization ID "100000" is "customer support center" as shown in Figure 4(a) which associates organization IDs with organization names. The active structure ID "200000" is "sales representative" as shown in Figure 4(b) which associates active structure IDs with active structure names. The behavior ID "200000" is "complaint handling" as shown in Figure 4(c) which associates behavior IDs with behavior names. The passive structure ID "200000" is "complaint information" as shown in Figure 4(d) which associates passive structure IDs with passive structure names. The relationship ID “2” corresponds to “customer support center”, “sales representative”, “complaint handling”, and “complaint information” in the evaluation target organization 3 .

The similarity data 13 is data that associates the evaluation with the similarity between the relations of the evaluation target organization 3. The similarity is calculated by the similarity calculation unit 23 described below. The similarity data 13 shown in FIG. 6 associates, for each evaluation, the relation ID with the highest similarity with the similarity between that relation ID.

The acquisition unit 21 acquires an evaluation for the evaluation target. The acquisition unit 21 acquires an evaluation for the evaluation target organization 3 from the evaluation providing device 2. The evaluation providing device 2 is, for example, a posting site such as a microblog, an SNS (Social Networking Service) site, a word-of-mouth site, the company's own CRM system, etc. The acquisition unit 21 may acquire multiple evaluations. The acquisition unit 21 may acquire the evaluation from the site or system in a pull type, or in a push type.

The acquisition unit 21 may acquire evaluations with negative emotion scores from among the evaluations of the evaluation target. The acquisition unit 21 may calculate an emotion score for each evaluation of the evaluation target, and acquire negative evaluations with emotion scores below a predetermined value. In an embodiment of the present invention, a higher emotion score indicates a more positive impression of the evaluation target organization 3, and a lower emotion score indicates a more negative impression of the evaluation target organization 3. The acquisition unit 21 calculates the emotion score for each evaluation, for example, using the Google Natural Language API, and filters out evaluations with emotion scores below -0.7.

The acquisition unit 21 may acquire an evaluation corresponding to the name of an entity from among the evaluations for the evaluation target. For example, the acquisition unit 21 may use the organization name, active structure name, behavior name, and passive structure name of each table shown in FIG. 4 as search keys to acquire an evaluation that matches the search key. The search key may be the name of an entity corresponding to any of the structures of the organization layer, active structure layer, behavior layer, and passive structure layer. The search key may be the name of an entity that satisfies a predetermined condition. The search key may be a name selected by the worker from among the names of each entity.

The acquisition process by the acquisition unit 21 will be described with reference to Figure 7. Note that the process shown in Figure 7 is an example and is not limited to this.

First, in step S101, the acquisition unit 21 collects evaluations that match the search word. In step S102, the acquisition unit 21 calculates an emotion score for each evaluation acquired in step S101.

In step S103, the acquisition unit 21 filters the evaluations whose emotion scores calculated in step S102 are equal to or less than a threshold value. In step S104, the acquisition unit 21 stores the evaluations filtered in step S103 in the evaluation data 11.

The definition unit 22 generates model data 12 for the evaluation target organization 3. For example, when an operation model is created using ArchiMate, the definition unit 22 obtains the output result and generates the model data 12.

The similarity calculation unit 23 calculates the similarity between each relation of an entity of the operation model in the evaluation target and the evaluation. The similarity between a relation and an evaluation is the similarity between the composite vector of the entity name of the relation and the composite vector of the words included in the evaluation. When the acquisition unit 21 acquires multiple evaluations, the similarity calculation unit 23 calculates the similarity between the evaluation and each relation for each of the multiple evaluations.

The composite vector is calculated, for example, using the Python gensim library and Word2Vec, which uses a neural network. The composite vector of the relation entity names is calculated by processing the relation entity names with the gensim library and Word2Vec. Furthermore, the composite vector of the ratings is calculated by processing the words included in the ratings with the gensim library and Word2Vec. Here, frequently occurring words in the ratings may be excluded.

The similarity between the relation composite vector and the evaluation composite vector is calculated by any method. For example, it may be Cos similarity, may be calculated by Doc2Vec, or may be calculated by a combination of these methods.

The similarity calculation unit 23 stores the similarity of each relation and each evaluation in the similarity data 13. Alternatively, as shown in FIG. 6, the similarity calculation unit 23 associates an evaluation with the ID of the relation having the highest similarity for that evaluation and stores the association in the similarity data 13.

The similarity calculation process by the similarity calculation unit 23 will be described with reference to Figure 8. Note that the process shown in Figure 8 is an example and is not limited to this.

In step S201, the similarity calculation unit 23 converts each evaluation in the evaluation data 11 into a vector. In step S202, the similarity calculation unit 23 converts each relation in the model data 12 into a vector.

In step S203, the similarity calculation unit 23 calculates the similarity between the vector of each evaluation calculated in step S201 and the vector of each relation calculated in step S202. In step S204, the similarity calculation unit 23 associates each evaluation with the relation ID having the highest similarity, and stores the association in the similarity data 13.

The output unit 24 outputs, from among the relations, the relation that has a high similarity to the evaluation. For example, the output unit 24 outputs, for each evaluation, the relation ID that has the highest similarity to the evaluation. The entity identified by the relation ID that has the highest similarity to the evaluation responds by referring to the evaluation.

The output unit 24 may output, for each relationship, the number of evaluations for which the similarity is equal to or greater than a predetermined value. For example, as shown in FIG. 9, the output unit 24 outputs, for each relationship, the number of evaluations for which the relationship is determined to have the highest similarity. Since relationships with a large number of evaluations have many issues in improving the evaluations, negative evaluations can be effectively reduced by improving relationships with a large number of evaluations.

The output unit 24 outputs, for each relation, an index corresponding to the emotion score of the evaluation whose similarity is equal to or greater than a predetermined value. For example, for each relation, the output unit 24 calculates and outputs an index that has a positive correlation with the emotion score in the evaluation in which the relation is determined to have the highest similarity. The index is, for example, the average of the emotion scores of each evaluation, or the most negative emotion score among the emotion scores of each evaluation. Negative evaluations can be effectively reduced by improving relations related to evaluations that strongly express negative emotions.

The processing device 1 according to the embodiment of the present invention can match evaluations and evaluation targets through computer processing, which can significantly reduce the cost of evaluation improvement. In addition, the relationship between the evaluation and the relations in the evaluation target organization 3 can be indexed as a similarity, so that the evaluation target organization 3 can efficiently work on evaluation improvement.

In the embodiment of the present invention, a case has been described in which the output unit 24 of the processing device 1 outputs relations that are highly similar to the evaluation. In the first and second modified examples, a case will be described in which improvement measures for relations that are highly similar to the evaluation are output.

(First Modification)
An output unit 24a according to a first modified example will be described with reference to Fig. 10. The output unit 24a refers to enterprise architecture data (EA data) 111 that accumulates improvement results, and acquires and outputs data related to relationships that have a high similarity to the evaluation from the EA data 111. The data output here can be improvement measures for relationships that have a high similarity to the evaluation.

The output unit 24a includes EA data 111, recommendation data 112, an identification unit 121, relationship data 122 and a recommendation unit 123.

EA data 111 is enterprise architecture data. For example, as shown in FIG. 11, EA data 111 has the following items: EA identifier, As-Is, To-Be, and Union-EA. An identifier that specifies the relationship between As-Is, To-Be, and Union-EA is set in the EA identifier item. In the As-Is item, data of the As-Is model of an enterprise architecture model in which other companies have realized improvements is stored. In the To-Be item, data of the To-Be model of an enterprise architecture model in which other companies have realized improvements is stored. In the Union-EA item, data of an integrated model of an enterprise architecture model in which other companies have realized improvements is stored. The integrated model is a model in which the As-Is model, To-Be model, and Transition are linked. In the first modified example, a case will be described in which EA data 111 is generated in advance.

The recommendation data 112 is data that identifies improvement measures output by the recommendation unit 123. The recommendation data 112 may be, for example, an EA identifier of the EA data 111, or data of each item identified from the EA identifier.

The identification unit 121, similar to the processing of the output unit 24 according to the embodiment of the present invention, outputs, among the relations of the entities in the evaluation target, the relations having a high similarity to the evaluation as the relation data 122. The relation data 122 only needs to be convertible into the entity name related to the relation, and may be the relation ID having the highest similarity to the evaluation, or each entity name converted from the relation ID having the highest similarity to the evaluation.

The recommendation unit 123 refers to the EA data 111 to identify an EA identifier related to the relationship data 122. The recommendation unit 123 searches the EA data 111 using the name of each entity identified from the relationship data 122 as a search key, and identifies an EA identifier similar to the search key. The recommendation unit 123 outputs the identified EA identifier, or data of each item identified from the EA identifier, to the recommendation data 112.

In the first modified example, the output unit 24a outputs, as a proposed improvement measure, an enterprise architecture model related to the relationship that is currently being requested to be improved, based on past improvement measures by other companies, etc. Among the output improvement measures, improvement measures that have solved a similar problem to the current one can be referenced. This allows the output unit 24a to output a relationship that is highly related to the evaluation, and to quickly find a solution to the relationship.

(Second Modification)
The output unit 24b according to the second modified example will be described with reference to Fig. 12. The output unit 24b sets the data of the As-Is model of the EA data (enterprise architecture data) 111 as input data for the encoder, learns the prediction model data 113 using the data of the To-Be model of the enterprise architecture data 111 as input data for the decoder and correct answer data, and sets the data of a relationship having a high similarity to the evaluation as input data with reference to the prediction model data 113, and outputs the data of the To-Be model. The data of the To-Be model is considered to be the future of a relationship having a high similarity to the evaluation, and can be an improvement measure.

The output unit 24b includes EA data 111, recommendation data 112, predictive model data 113, an identification unit 121, a recommendation unit 123a and a learning unit 124.

The EA data 111, recommendation data 112, identification section 121 and relationship data 122 are the same as in the first variant example.

The learning unit 124 learns using the EA data 111 as input and outputs the prediction model data 113. The learning unit 124 uses the EA data 111, which has already been used to improve business operations, to construct a Seq2Seq (sequence to sequence) model using teacher coercion. As shown in FIG. 13, the As-Is model data of the EA data 111 is set as the input of the encoder unit. The To-Be model data of the EA data 111 is set as the input of the decoder unit and as the correct answer data. The As-Is model data and To-Be model data of the EA data 111 are written, for example, in XML (Extensible Markup Language) format.

The learning unit 124 tunes an optimization algorithm or the like, and generates prediction model data 113 using a model in which the error has sufficiently converged. The learning unit 124 learns a prediction model that predicts a To-Be model from an As-Is model of EA data. When the learning unit 124 learns using, for example, Python's Keras framework, the loss value obtained as a result of learning becomes the error. When the loss value is infinitesimally close to 0, it is determined that the error has sufficiently converged.

When the learning unit 124 generates the prediction model data 113, the recommendation unit 123a inputs the relations in the relationship data 122, specifically the entity names of relations that have a high similarity to the evaluation, as an As-Is model into the prediction model data 113. The recommendation unit 123a acquires and outputs a To-Be model as an improvement measure for the relations identified by the identification unit 121.

In the second modified example, the output unit 24b refers to the prediction model data 113 that has learned past improvement measures by other companies, etc., and outputs a To-Be model related to the relationship that is currently required to be improved as a proposed improvement measure. This allows the output unit 24b to output a relationship that is highly related to the evaluation, and to quickly find a solution to the relationship.

The processing device 1 of the present embodiment described above is, for example, a general-purpose computer system including a CPU (Central Processing Unit, processor) 901, a memory 902, a storage 903 (HDD: Hard Disk Drive, SSD: Solid State Drive), a communication device 904, an input device 905, and an output device 906. In this computer system, the CPU 901 executes a processing program loaded on the memory 902, thereby realizing each function of the processing device 1. A GPU may also be used in conjunction with the CPU 901.

The processing device 1 may be implemented in one computer or in multiple computers. The processing device 1 may also be a virtual machine implemented in a computer.

The processing program of the processing device 1 can be stored on a computer-readable recording medium such as a HDD, SSD, USB (Universal Serial Bus) memory, CD (Compact Disc), or DVD (Digital Versatile Disc), or can be distributed via a network.

The present invention is not limited to the above-described embodiments, and many variations are possible within the scope of the invention.

REFERENCE SIGNS LIST 1 Processing device 2 Rating providing device 3 Rating target organization 11 Rating data 12 Model data 13 Similarity data 21 Acquisition unit 22 Definition unit 23 Similarity calculation unit 24 Output unit 111 EA data 112 Recommendation data 113 Prediction model data 121 Identification unit 122 Relation data 123 Recommendation unit 124 Learning unit 901 CPU
902 Memory 903 Storage 904 Communication device 905 Input device 906 Output device

Claims (7)

a storage device that stores model data for defining an operation model of an organization to be evaluated, the model data associating identifiers of a plurality of entities selected from each of a plurality of hierarchies forming the organization to be evaluated with identifiers of relations;
an acquisition unit that acquires a rating expressed in a language for the rating target organization from a rating providing device ;
a similarity calculation unit that refers to the model data and calculates a similarity for each relation from a composite vector of a plurality of entity names corresponding to each of identifiers of a plurality of entities associated with an identifier of the relation and a composite vector of words included in the evaluation;
and an output unit that outputs, from among the relations, a relation that has a high similarity to the evaluation. The acquisition unit acquires a plurality of evaluations,
the similarity calculation unit calculates a similarity for each relation of the plurality of evaluations from a composite vector of an entity name corresponding to an identifier of an entity associated with an identifier of the relation and a composite vector of words included in the evaluation;
The output unit outputs, for each of the relations, the number of evaluations for which the degree of similarity is equal to or greater than a predetermined value.
The processing device of claim 1 . The acquisition unit acquires a plurality of evaluations and calculates an emotion score for each of the plurality of evaluations;
the similarity calculation unit calculates a similarity for each relation of each of the evaluations filtered by the emotion score from a composite vector of an entity name corresponding to an identifier of an entity associated with an identifier of the relation and a composite vector of words included in the evaluation ;
the output unit outputs, for each of the relations, an index corresponding to an emotion score calculated for an evaluation in which the degree of similarity with the relation is equal to or greater than a predetermined value ;
The index is the most negative sentiment score among the sentiment scores of each evaluation.
The processing device of claim 1 . The output unit acquires data related to an entity name corresponding to an entity identifier associated with an identifier of a relation having a high similarity to the evaluation from enterprise architecture data that accumulates improvement results , and outputs the data.
4. The processing apparatus according to claim 1. The storage device further stores enterprise architecture data that associates data of an As-Is model and data of a To-Be model of the enterprise architecture model in which the improvement is realized;
The output unit sets data of an As-Is model of enterprise architecture data as input data , learns prediction model data using data of a To-Be model of the enterprise architecture data as correct answer data , refers to the prediction model data, sets data of an entity name corresponding to an identifier of an entity associated with an identifier of a relationship having a high similarity to the evaluation as the input data, and outputs data of the To-Be model.
5. The processing apparatus according to claim 1. A computer stores model data for defining an operation model of an organization to be evaluated, the model data associating identifiers of a plurality of entities selected from each of a plurality of hierarchies forming the organization to be evaluated with identifiers of relations;
The computer acquires, from a rating providing device, a rating expressed in language for the rating target organization ;
the computer refers to the model data and calculates, for each relation, a similarity from a composite vector of a plurality of entity names corresponding to each of the identifiers of a plurality of entities associated with the identifier of the relation and a composite vector of words included in the evaluation ;
The processing method includes a step of outputting, by the computer, a relation having a high similarity to the evaluation from among the relations. A processing program for causing a computer to function as the processing device according to any one of claims 1 to 5.

Images